ftvalentini

set.seed(33)

x = runif(100, min=0, max=1)

outlier_values = c(1e2, 1e3, 1e4, 1e5, 1e6)
for (i in outlier_values) {
  x[100] = i
  y = (x - mean(x)) / (sd(x))
  print(y[100])
}

ftvalentini

import numpy as np

n = 1 # number of observations
k = 5 # number of categories
c = 1 # 2da categoria
dim = 3 # dimension of embeddings

E = np.random.rand(k, dim) # embedding matrix

x_onehot = np.full((k, n), 0) 

ftvalentini

import numpy as np
import torch
import torch.nn.functional as F

torch.manual_seed(99)

# 5 features, 3 clases, 1 ejemplo
X = torch.randn(5)
W1 = torch.randn(3,5)
W2 = W1.detach().clone() # una matriz de perdida por loss function

ftvalentini

z_scale = function(x) (x - mean(x)) / (sd(x))
rob_scale = function(x) (x - median(x)) / (IQR(x))

set.seed(12)
x = runif(10)
y = x
y[10] = 10 # datos con outlier

# sin outlier
x_zscale = z_scale(x)

ftvalentini

import numpy as np

A = np.arange(12).reshape(4,3)

# opcion 1
res1 = np.apply_over_axes(np.cumsum, A, axes=[0,1])
# opcion 2
res2 = A.cumsum(0).cumsum(1)
# opcion 3
res3 = np.cumsum( np.cumsum(A, axis=0), axis=1)

ftvalentini

import numpy as np

A = np.array([1,2,35,8])
n = 2
# idx of top n values (NOT SORTED)
idx_top = np.argpartition(A, -n)[-n:] 
# sort idx from largest value to lowest
idx_top = idx_top[np.argsort(-A[idx_top])] 
values_top = A[idx_top]

ftvalentini

# simulate highly correlated x1-x2
set.seed(8)
X = MASS::mvrnorm(
  n = 100
  , mu = c(0, 0)
  , Sigma = matrix(c(1, 0.99, 0.99, 1), nrow=2, byrow=T)
  , empirical = F
)
# simulate y
df = data.frame(

ftvalentini

import numpy as np
import scipy.stats as stats

arr = np.array([0,2,3,0,236,8])

arr_ranks1 = np.searchsorted(np.sort(arr), arr)
arr_ranks2 = stats.rankdata(arr, "average")
arr_ranks3 = stats.rankdata(arr, "average") / len(arr)

ftvalentini

# when using yield in a function, you are creating a generator object 
def cuadrado():
    for i in range(10):
        yield i**2
        
for num in cuadrado():
    print(num)